Slashdot Mirror

A computer is something that can compute things. It shouldn't matter whether the box contains one THz chip, or a million Vic-20's, or smart goblins. If it can solve a dense system of linear equations quickly, then you've got something.

If you'd try this benchmark on distributed.net, you'd get abysimal performance. So if you view distributed.net as one computer, the benchmark will tell you that as a whole it isn't a fast *computer* (maybe that's what you had in mind). But this Japanese machine probably is.

Yes !

You're absolutely right. French -> english very faulty translation module.

My point is that it's a very _big_ number for beowulf clusters. The biggest one have 8192 processors ( http://www.top500.org/list/2001/11/)
and the alimentation and network problem are huge for Beowulf of this size.

What for 3.2 millions of processors ?

So are you saying that a cluster isn't a supercomputer? Check out the Top 500 list. Notice how many Crays are in the top 15? ASCI White is a cluster and is definitely a supercomputer. Clusters are the future of supercomputing.

• #25 - 795GFlops - Charles Schwab
• #40 - 618GFlops - Real World Computing Project http://pdswww.rwcp.or.jp/ - seems to be a computer industry consortium
• #46 - 648GFlops - State Farm Insurance
• #47 - 546 GLops - Saudi Aramco
• #49 - 536 GFlops - Rottendorf Pharma GMBH
• #53 - 512 - State Farm again
• #54 - 507 - Compaq
• #62 - 447 - Cray
• #69 - 441 - Financial Institution, Hong Kong
• #71, #72 - 420 - "Service Provider"
• #73 - 420 - Sun
• #78 - 371 - IBM TJWatson
• #95 - 315 - Silicon Graphics
• #99 - 301 - Edinfor, Portugal http://www.edinfor.pt/

Let's have a look at this list, shall we?

Do you see India or Pakistan anywhere on the top 20? No? How about any other 'rogue nation?'

Never mind that they'd need a bunch of highly classified test data to run simulations with.

I think we can safely say neither India nor Pakistan will be simulating any nukes, for the time being.

You can't.

As far as I know, the ASCI White system is pretty much "classified" in a way. I know it uses Power3 processors... over 8000 of them, in fact. And it has a mesh-type interconnect... very expensive stuff here. You can see how it compares here

Note that the second most powerful supercomputer peaks at HALF the GigaFlops... there's some serious power there. Though that list is technically innacurate... the #2 machine only has 3000 processors, and 8 on standby in case of hardware failure. That one is an "open" cluster, however. I don't think it's opened quite yet, and I can't even remember the name... something French... LeMue or somesuch.

I did take a look at it once, though. Sweet machine(s) :-) And you can't help but be impressed when you learn that it took 4.5 TONS of cabling to wire that puppy!

Actually, it's just as difficult to find the Linux machines on there...

But, most of them are listed Here with all of of the NOW Clusters ... The first one listed (#20) is a Linux Cluster

What's even more interesting is the fact that

The only Windoze cluster on the list (#320...as stated before) is the only cluster with that fast of processor per box (1 GHz) and that many installed processors (252) that low on the list.

Lets not forget that M$ has given up on all but the x86 (no more Alpha versions of their OS), yet it is hard to find any architecture that Linux will not run on.

However, this points to a new requirement for the top 500 list...that should be an OS column in this list...

Actually, it's just as difficult to find the Linux machines on there...

But, most of them are listed Here with all of of the NOW Clusters ... The first one listed (#20) is a Linux Cluster

What's even more interesting is the fact that

The only Windoze cluster on the list (#320...as stated before) is the only cluster with that fast of processor per box (1 GHz) and that many installed processors (252) that low on the list.

Lets not forget that M$ has given up on all but the x86 (no more Alpha versions of their OS), yet it is hard to find any architecture that Linux will not run on.

However, this points to a new requirement for the top 500 list...that should be an OS column in this list...

Then, point out the scads of Beowulf clusters and Linux/Unix based systems.

Finally, inform the rep and your management that you've chosen to use the more cost effective, higher performance and standardized choice...Unix.

If management resists further, do a cost analysis. That'll convince them.

299,792,458 m/s...not just a good idea, its the law!

I maintain a ms cluster which is a high availablity cluster. It runs exchange and works well. In looking for computational clusters, I found Top 500 cluster list. Most of the machinces are none intel based and can't run a ms cluster.

Alliance Supercluster Ranked as the World's Most Powerful Intel-based NT Cluster is a press release showing that a NT based cluster broke into the top 500 list in 207th place.

The microsoft site lists the cluster toolkit as a tech preview. It look to me that MS have found a new source of revenue and is now trying to get people to use it.

Quantum computing will, of course, take over everything. Assuming they iron out all the kinks and make it a marketable product- which could easily still be ten years off. But the potential is absolutely mind-boggling. The acceleration in computing power will make Moore's law obsolete.

Consider the speed at which you can compute things with an abacus, and the speed of the ASCI White, the world's fastest machine, running at about 7.2 teraflops. The difference in speed between these two things is orders of magnitude smaller than the difference between a relatively simple quantum computer and the ASCI White.

If you are thinking *very* long term it might be a good idea to start learning the new programming rules of quantum computing.

It's possible that updates to the Itanium (McKinley and so on) may prove to be more useful than the Itanium itself. Anyone remember when the 60MHZ Pentium was released? It was hot, expensive, huge and completely new, just like the Itanium. It wasn't really worth the purchase price, in comparison to what was available in 486 at the time. You have to start somewhere...

64bit CPU's will provide a much better range of addressing, both for RAM and filesystems. They will also be faster than their 32bit cousins. How do I know? Well, we have 2 large-scale linux clusters here at NCSA. The IA32 cluster has 512 dual PIII nodes at 1.0 GHZ. The IA64 cluster has 160 dual Itaniums at 800MHZ. The Itanium cluster is about 10% faster for its top500 run. Check out entries 34 & 41. Yes, the Itanium cluster has more RAM per node than the IA32 and it was running RedHat 7.1 versus RedHat 6.2 for its older sibling, but you're still looking at a factor of 3 in CPU to CPU comparisions.

To sum up, Itanium doesn't quite suck :) Let's hope the AMD 64bit CPU's are even better!

Also: Athlon and Pentium 4 block diagrams for comparison. That P4 is one ugly S.O.B....

Guys.....GUYS... listen: the MIPS chips are *not even close* to the same architecture as PowerPC or Intel/AMD.

The R14000 clock rates are expected to go up shortly, and at 500MHz the theoretical peak performance is 1 GigaFlop per second. They are much - much nicer chips than the mutts we run today.

Yes, these systems are not sometimes the best for handling vectorizable jobs, but they are so inexpensive compared to the old specialized hardware that it is easier to waste cycles than build special hardware.

As to memory bandwidth. Modern CPU caches make the question nearly moot.

If all of this were not true, then people wouldn't be building clusters and the majority of the top500 list wouldn't be dominated by clusters. Instead there are 3 traditional architecture machines in the top 20. This is the reason that Cray (etal) no longer dominates the marketplace... commodity systems have overtaken nearly all of the specialized hardware world.

-- Multics

There are a number of Beowolf clusters produced by IBM for their customers. I helped build (build = carried servers into room and racked) one setup for in New Mexico.

However, if you look at the Top 500 Clusters Site You will see many a IBM Linux cluster on there. Specifically numbers 41, 106, and 125. There are numbers more that aren't listed on that site that are going into many a commercial account every month.

What I am really wondering is: is there at the current moment ANY company/application/whatever that required this amount of storage? I thought that even a large bank could manage with a few TB's
Not intended as a flame, just interested

I work at a large credit card bank (we're the largest issuer of VISA cards, and our analytic data store is in the top 500 supercomputer sites). Our main Oracle data warehouse has about 38 TB of tablespace in use. It'll be awhile before we need drives with PB capacity. :-)

FYI, this will be updated after supercomputing, but this is the list of the fastest computers in the world.

The Pittsburgh's super'puter will rank up there with LANL's new one (also a Compaq based one). Pittsburgh's will be the fastest SC for nonclassified work.

I'm not sure whether or not it'll dethrone LLNL's ASCI White or not. It does knock seaborg @ NERSC from the fastest unclassified SC spot though.

Me too.

Some minor tidbits:

• Yes, LLNL does have at least one of the ``big bertha'' displays.
• We also have a lot of computing power (including the number 1 and number 4 machines on the current Top500 supercomputers list.

[Disclaimer: I'm a sysadmin with VIEWS (and we have an open sysadmin position).]

The cluster is at #385

A related site, which I find a bit more interesting, is the clusters database. Particularly noteworthy are three PC clusters that cross the teraflops line (peak performance, mind you, but still impressive).

Visit here to view 500 fastest computers in the world as of June 2001. Cray is actually number 11. IBM ASCI White SP Power 3 is the king.

It's interesting to note that a beowulf cluster is also there (#42)

The machine in Britain would barely rank #48 on top500.org, so what's your point?

I hate it when the press makes it sound like America is the jack-ass backwards donkey of the supercomputing world. This writer implies the Japanese and Europeans have vastly superior computing power. This is clearly the notion of a chucklehead. Take a look at The Top 500. By its (Linpack) metric, 8 of the top 10 machines are in America. Three of them are DEDICATED to weather or environmental work (Naval Oceanographic Office, National Centers for Environmental Prediction). A fourth one at NERSC is relatively open, compared to defense machines, and I'd be willing to bet weather code is running on it regularly. These are all teraflops machines. Japan has the other two in the top 10. Anybody know the job mix on those two? Europe's fastest machine is the Hitachi in Muenchen. The fastest dedicated European weather machiens are the T3Es at the Deutscher Wetterdienst and at the UK Meteorological Office.

I don't buy these whiny weathermen's complaints. The difference is that the American machines are all massively parallel machines (mostly IBM SP). The Japanese manufacturers all make vector machines, some of which the Europeans use. The Cray T3E is kind of a weird in-between architecture. It takes a good programmer to use a MPP to its full capability. The vector users, on the other hand, have 30 years of old code and practice which keeps them in the game. If the Americans would suck it up and learn to use their amazingly fast IBMs we would hear whining from the other side of both ponds. If you try to run your old code for the Cray C90 on an IBM SP, you are going to get terrible performance. If you rewrite the code, you may get great performance. But these guys aren't rewriting the code. Take for example the machines at NCEP. These create the daily production weather models used all over the US. They are IBMs which replaced a Cray that self-immolated about 1.5 years ago. When they brought the new machines up, I wonder if they rewrote the code beyond making it run? If you know, enlighten us!

Some of these have also been noted by other respondants, but let me see if I can summarize:

• Motorola Power PC: With the next generation of faster G4 processors comming out and Apple's continuing commitment to the architecture, I wouldn't write this one off yet.
• IBM Power PC: I haven't exactly figured out what diferentiates IBM's and Motorola's PPC chips, but it seems like IBM's are targeted more towards embeded applications where they'll probably remain a big player for some time.
• IBM Power: It seems with the IBM - Motorola split on the PPC, IBM returned to the power architecture for their big boxes. The latest offering is the Power III, which gives the Alpha and Itanium a serious run for their money. Keep in mind that this processor is at the heart of the world's fastest computer, so I think it's providing some stiff compition for Intel. (Disclaimer: I may have some bias given that ASCI White is just a few blocks away from me right now.)
• MIPS: I think I heard that SGI is migrating all their MIPS machines to Itanium, so this one's probably dead.
• PA-RISC - HP worked jointly with Intel on Itanium with the intention of it replacing PA-RISC, so I think it safe to declare this one out of the running.
• AMD's x86-64 architecture - I honestly don't know much about this architecture, but from my understanding it's a derivitive of the Intel x86 architecture with extentions for 64 bit memory addressing. Whether or not you want to consider it an Intel design is up to you, but Intel certainly isn't going to endorse them and I think they'll provide some stiff competition for Itanium (particularly if they run existing x86-32 code flawlessly).

As for some of your other comments:

• With the new higher speed UltraSparc III processors that Sun is putting in their new Blade machines, Sun has once again reached a reasonable price/performance point, so I wouldn't discount them too much yet. I completely don't understand your comment about the ESA not being able to afford it. Furthermore, Sun is very anti-Intel: I've heard the only place in Sun that they're allowed to have Intel boxes is is the Solaris x86 development group, so Sun will likely be the last major workstation/server vendor to make the switch to hop on the Intel bandwagon.
• Motorola most certainly still makes processors. Besides the afore mentioned PPC, the 68k architecture is still around since it's great for embedded applications and handheld computers, such as the Palm (which last I heard used the Dragonball processor - a variation of the 68040, I think).

In summary, I think your conclusion that the Intel based design is the only serious contender out there is a bit overstated.

-"Zow"

MayorQ

here! Enjoy!

Maybe I'm missing something, but when someone says Cray is the "only mainstream recognizeable name in supercomputing" I think they're seriously out of touch. At least out of touch with the list of Top 500 supercomputing sites.

IBM has at least half the entries here. And the all important #1. Cray shows up at #10... after 5 IBM installations.

What's up? Is someone claiming Cray belongs to some special set of supercomputing? Would they care to elaborate? Or am I just being overzealous....

-A

had the good fortune to be playing with this recently via simulation. If you give the processor a *huge* instruction window (256 instructions) and the ability to execute *any* number of instructions of *any* type in parallel (except for memory accesses - see below), you still get an average Instructions Per Clock of about 2.1-2.2. 95% of the time, you're getting four instructions or fewer issued (and most of the time, far fewer than that).

Yes. You average 2.1-2.2, and 95% of the time you're only getting 4 or fewer. However, when you look at the other stuff the Power3 architecture includes, it's pretty obvious what the overall intent is:

Hardware Loop Unrolling.

IBM has got some customers that use some serious CPU. We're talking national labs and the like. For them, the ability to run 8 of those neat 'multiply-and-add' instructions per clock cycle is quite an important feature.

The chip *starts* at 375mz, and can do 16 floating point ops/clock (an amazing amount of code uses that mult-and-add over an array - and the IBM compilers are smart enough to detect and convert divide to multiply-by-inverse and add/subtract issues).

And of course, IBM is hoping that even though the big SP/2 iron is limited to national labs and Fortune-500 companies (see The Top500 List for details), that they'll be able to sell a lot of the smaller 43P deskside boxes (1-4 Power3 CPUS) and the 8-16 CPU rackmount servers, to all the smaller companies that need number-crunching.

This isn't going to work. Distributed computing over 56K dialup modems is simply commercially infeasible.

Why? Because an embarassingly parallel commercially interesting problem is an oxymoron. The interesting problems are ones which require inter-node communication of at very least tens of MB/day.

Computer scientists have known this for decades. That's why linpack is used as a supercomputing benchmark, while RC5 is not.

The interesting problems are the hard ones, the ones which aren't embarassingly parallel. The ones which can't run off of a 56K dialup modem.

I thought this was kinda funny...

Global supercomputer leader Cray Inc

Gee, Cray is definitely the "Global supercomputer leader." Their fastest computer is an incredible .18 times as fast as IBM's.

[According to the latest TOP500 Supercomputer Sites list, Cray's fastest computer is ranked tenth.]

-thz

Have we suspended Moore's law? The ASCI Red (currently 2nd on the Top 500 list) has 9632 processors, and it broke the Teraflop boundary with plain ol' Pentium Pro processors--1996 technology. (It's since been upgraded) Now, say this vaporware computer shows up 9 years later, that will give processor performance roughly 2^6 of an improvement. With 10-20K processors, that's roughly 64-128 Teraflops, not a Petaflop.

Now, 100-ish Teraflops is a lot more realistic than a petaflop, especially since this article quotes a figure of 100, not 1000 Teraflops.

The www.top500.org only lists 1 NT system and 499 UNIX systems. And to keep things fair of the 5 "self-made" models there is 1 nt system and 4 UNIX systems. So, I guess that I'm not the only one who cannot tune NT properly.

On a related note, has anyone else noticed that the claims of IBM's death were a bit premature? They seem to really have re-emerged as a major leader in R&D, and are piling up market leadership points in hard drives and Java tech while making inroads into monitors, back into PCs, and regaining ground lost to Sun, Intel, SGI, and DEC in the big-iron and supercomputing markets.

I don't think IBM is "regaining lost ground" in supercomputing, I think IBM has nearly stolen the show. Take a look at the current Top 500 list of supercomputers and you'll find that IBM built the current champion as well as 2 of the top 3, 5 of the top 10 and almost half of the top 500.

Add to that all of the other cool R&D stuff, plus lots of nifty software innovations (Alphaworks, anyone?) and the largest consulting and professional services organization in the world and it's pretty clear that Big Blue is very healthy.

So, please, everyone go buy some IBM stock and drive up the value of my options ;-) (yes, I work for IBM, and it's a pretty cool place to work, at the moment).

This may be out of your price range, but Sandia Labs has a nice little machine they call the Cplant (Computational Plant). Its a cluster of about 500 linux boxes with supercomputer power. Its ranked 84th in the list of TOP500 Supercomputers in the world as of November 3rd.

More details about the architecture of this supercomputer can be found in Overview of Recent Supercomputers.

Mirror of the TOP500 Site

Anas

Wow, decent stuff posted in the middle of the night to /., the first post is not a troll, and this computer might still make it on to TOP500 by the time it exists :-)

-M5B

That is not really what this system is meant for (running one job on all of the processors). On our Origin2000's (we have 3), we run many "small" jobs through a queueing system. (Small = a handful of processors and a dozen GB of memory or so.)

Each of these jobs are larger than most jobs one would run on a cluster like the IBM SP (got a few of these) or a Linux cluster (got one in my lab, getting a big one soon), becuase they are memory bound rather than CPU bound. That is, it is easier to address 10GB of RAM than to fiddle with message passing between nodes and such.

P.S. "we" are msi.umn.edu, and you should be able to find at least one of our supercomputers on the Top500 site.

According to the 1998 review of a Sun Enterprise 10000 at UnixReview (previously Performance Computing)

Silicon Graphics Inc. (SGI) has published results for its Origin2000, a cache-coherent Non-Uniform Memory Architecture (ccNUMA) system, of 9,478 for SPECfp_rate95 and 5,922 for SPECint_rate95, both for system configurations with 64 195MHz Mips R10000 CPUs. The SGI floating-point score was on a system with 16GB RAM, while the integer score was obtained on a system with 48GB RAM. The SGI machine's floating-point performance is significantly higher than Sun's, while the Origin2000's integer performance is only slightly higher than the Starfire's. Additionally, SGI issued a press release on Feb. 25, 1998 announcing SPECrate scores for its latest 250MHz Mips R10000 CPU of 11,984 for SPECfp_rate95 and 8,021 for SPECint_rate95. At review time, however, those results had not yet been posted on the SPEC Web site, so configuration details of the test systems were not available.

Perhaps because only ASCI can afford those way too expensive computers?
I think when ASCI Red was unveiled, they allowed for up to 50% of the CPU time to be used for non-ASCI research projects.

The top of the line supercomputers are listed in TOP500 Supercomputing Sites latest list. Clearly there are more than ASCI Nuclear Research computers. I don't remember Slashdot reporting installation of the two new non-ASCI teraflop computers, though. Perhaps only the best is of interest to the slashdot crowd.

What's kinda funny is to do a little comparison of this machine computer to the others on the Top500 list. Right up in the top 10 are two Japanses machines (Hitachis, I think) BUT they only use ~100 processors, not ~8000 or ~9000. Dang impressive. Not only that, but all the Hitachi and Fujitsu machines seems to be much, much smaller. Curious -- does anyone know what the $/flop difference between the two kinds of computers is?
Zorn

[sarcasm]A terraflop? Wow, a couple of years ago I remembered hearing about a supercomputer that could barely do 5 terraflops. So i guess there should be couple G4 multi-proc systems on the Top500 list. Oh, wait, no, your just an idiot mac user who is slave to apple and cant get his flop-ratings right [/sarcasm]

I know im going to lose lots of karma for this, but this guy deserves it!

Mark Duell

Neither one of these articles provided any interesting details about the Cluster. What language are they using? How efficient is the message passing (is it PVM or MPI)? What kind of benchmarks does this thing have? Is it pushing GFLOPS or TFLOPS? Does it beat out everyone in The Top 500 List. I just think some details about how great this system is would be nice, instead of just a RedHat advertisement! I know RedHat must be proud that their distribution is being used, but I would like some more concrete information than some sugarcoated marketspeak.

sid=lug

Really? So you've done this?

I wonder if IBM is actually gonna write a NUMA layer for linux?

SGI has the best in the business, and they're busy putting it into linux as we speak. However, IRIX is _the_ OS as far as NUMA goes, coupled with the sgi Origin. With sgi's massive linux movement, I'm sure linux will be up to par with IRIX in no time at all. Check em out at www.sgi.com/origin. Ever heard of ASCI Blue Mountain? Checkout the top 500 list.

-rpl

What i think you are excluding in your price/performance analysis is the architecture under which the SGI operates. It will give IMMENSELY more performance using all 8 CPU's than 8 separate Athlon systems, especially when you have tasks which can use the same memory spaces and such. If you check the top 500 supercomputers list, SGI is basically the top supplier of high-end supercomputers (obviously owing to the fact that they bought Cray a few years ago). If you look closely, their top end systems with only 256 CPUs beat the shit out of systems with 1000's. As with all "Supercomputers" vs. clusters, you pay for the ability to intercommunicate b/w the cpu's at an increadibly high rate, way beyond normal pc architectures... so in this scenario, the CPU is not the bottleneck. BTW, to stay on topic, I would suggest the Athlon for you real-time computations (for reasons already stated - better fpu, faster core...) *anyone get the reference?*

it is on top500.org